fix(complete): bash dynamic engine passes tokenized $2 instead of raw cursor word

## Summary The bash glue script emitted by the dynamic completion engine (`CompleteEnv` / `write_registration` in `clap_complete/src/env/shells.rs`) uses `words[COMP_CWORD]="$2"` to capture the word under the cursor. This is incorrect: `$2` is bash's tokenized word, which word-splits on spaces. As a result, completion breaks for arguments that contain spaces. ## Reproduction Any CLI using `CompleteEnv` with an argument that can contain spaces — for example a `--reader` flag whose value is a reader name like `"Yubico YubiKey OTP+FIDO+CCID 00 00"`: ```bash $ mycli --reader "Yubico YubiKey <TAB> # completer receives "YubiKey" instead of "Yubico YubiKey " # completion fails or produces wrong results ``` ## Root cause The bash-4 guard in the generated script: ```bash if [[ "${BASH_VERSINFO[0]}" -ge 4 ]]; then words[COMP_CWORD]="$2" fi ``` `$2` in a completion function is the word under the cursor as tokenized by readline — it splits on whitespace. For space-containing arguments, this loses the portion before the last space. ## Fix Read `COMP_LINE` up to `COMP_POINT` and strip the last word-break prefix, exactly as the **static** bash generator already does since `clap_complete` 4.4.6: ```bash if [[ "${BASH_VERSINFO[0]}" -ge 4 ]]; then local _cur="${COMP_LINE:0:${COMP_POINT}}" _cur="${_cur##* }" words[COMP_CWORD]="${_cur}" fi ``` This fix was applied to the static generator (`aot/shells/bash.rs`) in 4.4.6 but was never ported to the dynamic engine (`env/shells.rs`). ## PR A fix is proposed in https://github.com/douzebis/clap/tree/fix-bash-dynamic-completion-words-compopt — PR to follow.